[PATCH] MCInstrAnalysis::evaluateBranch() silently fails without asserts

[Stephen Checkoway]

On Oct 16, 2013, at 5:34 PM, Ahmed Bougacha <ahmed.bougacha at gmail.com> wrote:

> On Wed, Oct 16, 2013 at 11:07 AM, Stephen Checkoway <s at pahtak.org> wrote:
>> MCInstrAnalysis::evaluateBranch() was assuming that if the OperandType is MCOI::OPERAND_PCREL, then the operand was an immediate. This appears not to be the case.
>> 
>> I'm unsure if this is handling it correctly or if we should just give up immediately if it's not an immediate or something else.
> 
> There was a patch attempt some time ago
> (http://thread.gmane.org/gmane.comp.compilers.llvm.cvs/148669), but
> this area didn't change a lot as far as I remember. The gist of it:

Glad to see I was on the right track (although my patch didn't do it quite right). Was there a reason that the patch you linked to didn't make it in?

S

> 
> - the problem is in the way -cfg and -symbolize interact. -cfg doesn't
> expect a symbolized disassembly. -symbolize enables the MC*Symbolizer
> in llvm-objdump's MCDisassembler. Eventually it will be very useful to
> handle symbolized MCInsts, but, as is, the easy (/hacky) fix would
> probably be to move the -symbolize block in llvm-objdump after the
> -cfg block.
> 
> - evaluating symbols in MCInstrAnalysis is the right thing to do, but
> it requires that the symbols created by the Symbolizer are assigned
> correct addresses.
> 
> Hope it helps, I'm sorry I can't spend more time on this.
> 
> - Ahmed
> 
>> I don't know how to write a test case for this since the assert only triggers in Debug.
>> 
>> --
>> Stephen Checkoway
>> 
>> 
>> 
>> 
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> 
> 

-- 
Stephen Checkoway